- Joined
- Oct 19, 2019
- Messages
- 7,737
- Reputation
- 7,911
- Reaction score
- 39,171
- Points
- 0
- Currently Smoking
- Various strains, via various intake tools, but no combustion.
Actually, the long term picture is even worse as AI gradually trains itself more and more on other AI output as more and more content available for training has already been generated by AI. There is a real possibility that the quality (if I can even use that term) of the output could decline to the point of uselessness due to this training problem.More specifically, it was trained on RANDOM INTERNET BULLSHIT. And to set the record straight, there is zero intelligence here. It is just an illusion, much like all the other attempts at simulating "Intelligence" over the decades.
There is no question that AI is going to change a lot, particularly when specifically designed and applied to particular task, such as, for example, medical imaging (combination of AI and expert checking already beats the best experts on their own). Much computer programming is now being done by AI followed by checking by a human programmer. I suspect that medical imaging has alreay been found spectacularly effective at other stuff. I;d bet that the underlying image analysis in both Merlin bird app and the INaturalist apps are repurposed image analysis originally intended for medical imagery.
But large language models trained on text on the internet are another matter altogether in the long run. The dudes who figure this stuff out have to figure out how to avoid the obvious training traps, or the whole shebang could come down in a bleeding pile.
